McKinsey and Bain on agentic AI
McKinsey reports two‑thirds of enterprises are experimenting with AI agents but fewer than 10% have scaled them, citing weak data foundations as a common limit. Bain recommends a federated data‑governance approach—structuring ownership across domain teams and layering data/knowledge, analytics, and orchestration—to move from experiments to reliable agentic systems. ( )
McKinsey’s State of AI 2025 finds 62 percent of organizations are experimenting with AI agents while no more than 10 percent of business functions report scaled agent use. (studylib.net) The McKinsey online survey ran June 25–July 29, 2025, gathered 1,993 responses from 105 countries, and reports 88 percent of respondents use AI in at least one business function. (studylib.net) McKinsey also reports 23 percent of organizations have scaled an agentic AI system in at least one function, but most of those deployments are limited to one or two functions. (forbes.com) Bain & Company recommends a three‑layer architecture for agentic AI—data and knowledge, analytics, and orchestration—and a federated data‑governance model that assigns ownership across domain teams. (bain.com) Bain warns that “security and governance must be embedded by design, not bolted on,” and it calls for centralized registries and policy layers to govern agent access and execution. (bain.com) McKinsey and other analysts identify weak or fragmented data foundations—missing schema contracts, inconsistent metadata, and siloed stores—as common limits that cause agents to break when moved from pilot to production. (studylib.net) (technologyreview.com) Bain says agents need unified, governed access to structured and unstructured sources (relational, vector, and graph stores) so persistent memory, tool catalogs, and multi‑step orchestration work reliably. (bain.com) McKinsey notes agent adoption concentrates first in IT and knowledge management, where use cases like service‑desk automation and research augmentation have matured. (studylib.net) Both firms point to organizational fixes—workflow redesign, executive sponsorship, and federated ownership of context and data—as the practical steps required to move from experiments to dependable, scaled agentic systems. (studylib.net) (bain.com)
